Lesson 1
Agents are workflows with autonomy
Testing must consider not just final text but intermediate plans, tool calls, retries and side effects.
All learning tracks
Agents, MCP & Future-ready EngineeringAdvanced 2 hrs
Testing Agentic AI
Test plans, tool selection, loops, permissions, side effects and recovery when AI can perform multi-step actions.
AgentsEvaluationSafety
Learning outcomes
Test agent behaviour
Detect unsafe loops
Validate tool-use boundaries
Guided lessons
Learn the engineering thinking
Lesson 2
Tool-selection accuracy
A correct answer produced through the wrong or overly privileged tool can still be an unsafe result.
Lesson 3
Loop and retry controls
Agents need bounded attempts and recovery behaviour to avoid runaway cost or repeated side effects.
Lesson 4
Permission boundary testing
Test whether the agent refuses actions outside the user, workspace or role permissions.
Lesson 5
End-to-end agent evaluation
Measure task success, safety, tool correctness, cost, latency and human intervention rate together.
Hands-on practice
Do something with what you learned
Do not just tick these mentally. Write the query, create the test matrix, refactor the code, or document the decision. Practical evidence is the point.